A while ago I had one of those “Someone on the Internet is Wrong” arguments with the authors of an article arguing that we would need massively more evidence before we could conclude that autonomous cars are safer than those driven by humans. Rather than dig back to find those arguments again, I thought I’d link to this Bloomberg piece and, in particular the following passage
GM’s autonomous test cars were in 22 accidents in California last year, according to data from the state’s Department of Motor Vehicles … In a November interview, GM President Dan Ammann attributed the accidents to testing in a dense urban environment and noted the company’s cars weren’t at fault in any of the incidents.
Suppose that in any crash between autonomous cars and humans, each is equally likely to be at fault. What is the probability of seeing 22 crashes caused by humans and none by autonomous cars. Obviously, it’s the same as that of a fair coin showing 22 heads in a row, which is 2^-22 or about 1 in 10 million.
Of course, the drivers involved in the crashes aren’t likely to be a random sample of the population. As is standard in such things, the 80/20 rule applies: 20 per cent of drivers are responsible for 80 per cent of crashes and traffic infringements. The 80/20 rule is derived from a Pareto distribution, and we can apply it a second time to say that 20 per cent of the remaining 80 per cent of drivers are responsible for 80 per cent of the remaining 20 per cent of crashes. That is, 4 per cent of drivers are responsible for 96 per cent of crashes. On that basis, it’s perfectly possible that the remaining 96 per cent of good drivers are as good as autonomous cars or even better.
It might also be argued that autonomous vehicles may fail in defensive driving, that is, in reducing harm in a crash caused by the failure of another driver.
Still, it seems pretty clear that autonomous cars are a lot better than the drivers responsible for most crashes and infringements. It isn’t that hard to identify a lot of these drivers before they kill themselves someone else, since prior driving record variables, particularly a driver’s prior traffic citation history, are the most consistent and powerful predictors of subsequent accident risk. Now that cars don’t need steering wheels or pedals any more, there’s no obvious reason to put people with bad driving records back in charge of them. Bad drivers should have their cars driven by robots.
{ 124 comments }
Frederic Bush 01.19.18 at 3:32 am
I don’t think you can keep iterating the 80/20 rule like that and expect it to hold true.
Mark Brady 01.19.18 at 3:49 am
And those of us who never learned to drive won’t need to learn and can be driven by robots!
Bill 01.19.18 at 6:00 am
Good drivers too, John.
CarlD 01.19.18 at 6:05 am
Who will buy them these cars?
SC 01.19.18 at 6:23 am
(Don’t tell Atrios.)
bad Jim 01.19.18 at 6:48 am
The most fearful, least competent drivers buy SUV’s for their putative safety, putting the rest of us at risk, and, almost as bad, slowing everything down as they attempt to navigate their enormous beasts down narrow streets.
SusanC 01.19.18 at 8:19 am
I suspect it will turn out that all humans are bad drivers. It may well happen that autonomous vehicles get sufficiently good that human beings are no longer allowed to drive on public roads.
Having said that, there may currently be a bad interaction between autonomous vehicles and bad drivers.
If I recall correctly, google’s cars suffer more rear end collisions that human drivers. As human drivers we mostly hate people tailgating, and try to take action to get away from the tailgater.
The autonomous vehicles possible suffer from an inteaction along the lines of: idiot human driver tailgates autonomous car. Autonmous car doesn’t know how to get out of this situation. Traffic lights ahead turn red. Autonomous vehicle hits the brakes. Idiot human driver runs into the back of autonomous car and its officially the humans fault for tailgating.
Ion D 01.19.18 at 9:49 am
I still don’t have my driving licence at 36, so, in a way, I would like to see autonomous cars in widespread use sooner rather later. However, I find Mr Quiggin’s argument very worrying: it’s almost a certain thing that at some point autonomous cars will drive better that any human. And, in a culture as obsessed with safety as the contemporary one, it’s also predictable that there will be those that will call for all the human drivers to be banned.
Ion D 01.19.18 at 9:50 am
@7
Sorry, I hadn’t seen that you made the same point.
Phil 01.19.18 at 10:23 am
In a November interview, GM President Dan Ammann attributed the accidents to testing in a dense urban environment and noted the company’s cars weren’t at fault in any of the incidents.
What evidential base could be sturdier than that?
Layman 01.19.18 at 10:31 am
Like Atrios, I’m deeply skeptical of most of the predictions about how driverless cars will be successful. Taxis that can’t wait double-parked at the curb and can’t help you with your luggage seem unlikely to replace taxis that can. Delivery vehicles that can’t actually deliver your packages to your door are not delivery vehicles. An article last week quoted industry touts as saying that driverless cars should have dedicated lanes, because the cars will drive more slowly and carefully – in fits and starts even – than normal traffic flow and will disrupt that traffic flow if not confined to dedicated lanes; but are dedicated lanes for personal single-occupant driverless cars a good idea at all? Why not driverless buses? Probably because the cost savings from eliminating bus drivers won’t fund the technology.
Aside from those practical objections, I remain deeply skeptical that they can be made to work in anything other than drastically constrained circumstances, anytime soon.
Driving is hard. It’s true people don’t do it well in some ways where computers will do it better – attention span, for example – but it is not at all clear that there are not aspects of it where humans will be better at it than AI for any foreseeable future. Anticipating the behavior of others, for example, as SusanC points out. Someone has just gotten into that car up ahead on the right curb. A human driver will anticipate that car will suddenly pull out into the street with no warning, because this is something that people do. Will the driverless car anticipate that? What will it do as a result? What driverless cars do now when they anticipate or detect a danger is stop dead in the street and wait for intervention, and there’s little reason to believe that will change in anything like the timescales proponents typically offer for adoption of the technology.
John Quiggin 01.19.18 at 11:24 am
@10 If you read the article, it links to the accident reports. I checked a couple, which showed the human driver at fault. You’re welcome to do a more thorough check.
@8 “The autonomous vehicles possible suffer from an inteaction along the lines of: idiot human driver tailgates autonomous car. Autonmous car doesn’t know how to get out of this situation. Traffic lights ahead turn red. Autonomous vehicle hits the brakes. Idiot human driver runs into the back of autonomous car and its officially the humans fault for tailgating.”
Not just officially but in reality, just as if the driver in front were human. Out of interest, how do you respond in this situation? Run the red light? Swerve into oncoming traffic?
Cranky Observer 01.19.18 at 11:28 am
General Motors treats its autonomous car program as a standard Big 3 development program and follows NTSB/NHTSA/State of California guidelines and reports all program accidents to the state in which they occur. Google/Uber/etc have been caught multiple times not reporting accidents, so we have to assume the Silicon Valley “car†companies vastly underreport.
Building cars at national scale, hard it is.
Ed 01.19.18 at 2:24 pm
One of the many big mistakes made with the whole car culture thing was the determination, if that was what that was, that everyone could and should operate a motor vehicle and that everyone (almost literally so in the US) would do so to get around.
I suspect this was prompted by the notion everyone knowing how to drive and to repair motor vehicles would be a big advantage in the next big war, though of course its pretty much impossible with new cars for their operators to repair them on by themselves, so you don’t even get that anymore. To this was added a recurring theme in the early automotive age, an almost willful blindness to see just how bad traffic was going to get.
Driving a car is actually pretty difficult and requires good co-ordination and motor skills, and constant attention, to do it properly. Bad driving is due to a failure in one of these areas. Any drugs or medication, medical conditions such as sleep apnea, just being under alot of stress or literally unable to concentrate can be fatal, often literally. This has been recognized and addressed -with a sledgehammer- only for recreational drugs.
If getting so many vehicles on the road was so important, they could have gotten that with much stricter licensing requirements, almost like getting a commercial license, but no taxi monopoly, so you would have a good chunk of the population not only very capable of driving, but willing and able to make some money on the side by driving the rest of the population, because they don’t make the cut for being competent drivers, around. Not only have you just created an entirely new center of economic activity, which capitalism should like, you’ve just eliminated almost all those accidents where bad driving was the primary cause. And the many people who don’t like driving now don’t have to.
The point is what they seem to be trying to do with a technological solution could have been done almost a century ago with human drivers, with some better organization.
JanieM 01.19.18 at 2:25 pm
I bought a brand new car a few years ago, the only time I’ve ever bought one new. I’m hoping to keep it for many years to come.
It’s been a great car for me. Except — the clock doesn’t keep the time correctly. It gains time, slowly but inexorably, until every few months it gets to about seven minutes fast and I set it back to one or two minutes fast. I have never had another clock in my life, inside or outside a car, that did that.
Every time anyone says driverless cars will be safer, I just laugh. They can’t even make the clock — the *clock*????it’s not like it’s some weird new invention that no one quite knows how to design — work right and they’re going to make cars safe mixed in with semis as big as trains, on icy roads in a Maine winter? Ha ha ha ha ha ha ha ha ha.
Ed 01.19.18 at 2:29 pm
To follow up on my earlier comment, a big change predicted by many people, though I’m not sure the auto manufacturers are on board with this, with automated cars is that individual car ownership would no longer be a thing. People would just call for an automated car to come and drive them when needed and not keep their own car sitting parked most of the time somewhere.
People now often half live in their cars, or sometimes completely live in their cars, and/ or use their cars for storage so this may be a cultural change too far (these are all bad driving habits but sometimes forced on people).
But the point is this scenario, of not having your own car but having a driver a phone call away to pick you up when needed, for a small fee, could have been the norm from the start with different regulations. I really think individual car ownership was promoted to get all those tank and truck drivers and mechanics for the next war.
steven t johnson 01.19.18 at 2:31 pm
CarlD@4 “Who will buy them these cars?”
Most likely, the hope is the bad drivers will be forced to buy them, after insurance companies raise their insurance rates to prohibitive levels to make driverless cars the only feasible alternative. A statute permitting insurance companies the administrative power to determine who is a bad driver, as opposed to interfering with free enterprise by requiring a legal procedure of some sort, may not even be necessary. Punitive rates are already permitted. Then if the state requires insurance, criminal charges can levied to force the changeover. Any money saved from lower accident rates will be appropriated by the insurance companies, barring price competition.
Glen 01.19.18 at 3:15 pm
I expect also that driverless cars will change the flow of traffic. Their presence will make some people better drivers and some people worse. It’ll be difficult to define a bad driver, for a while.
BruceJ 01.19.18 at 3:33 pm
Here’s the fly in your statistical soup: where’s your control group of human drivers? How do we know that there wouldn’t have been fewer (or more) than 22 accidents had both drivers been human?
BruceJ 01.19.18 at 3:40 pm
By keeping an an eye on, and managing, the idiot tailgater as well as watching the traffic in front of me. Start slowing down as I approach the intersection if it’s going to change before I get there (protip: watch the crosswalk signs, they now often feature a countdown telling the pedestrians how long until the light changes. ) Tap brakes a few times to flash the lights before I brake in earnest.
Defensive driving works.
Though sometimes I wish Uncle Albert’s Auto Stop and Gunnery Shop catalog was real :-P http://www.sjgames.com/car-wars/games/unclealbert2035/
afinetheorem 01.19.18 at 3:54 pm
Cars kill roughly 1.2 million people per year worldwide. Something like 90% of those are preventable. Quibbling about “who will deliver packages to your door” when we have literally invented a technology that will save more lives per year than a cure for malaria seems incredibly short-sighted.
There are a handful of places in the world that are quite laissez-faire with laws on self-driving cars. They are going to be home to a giant new industry. Those places where the status quo bias is so strong as to overwhelm the massive potential benefits here are the nutty ones.
(And on the calculation of good v. bad drivers – essentially everyone is bad when they begin learning how to drive. Conditional on one year of experience, I imagine self-driving cars are better than nearly every human driver. And I also imagine the learning curve is more rapid for AI than humans.)
L2P 01.19.18 at 4:41 pm
I don’t see the link to the article, but one thing to keep in mind is that our definition of “at fault” is pretty narrow. It generally looks only to the second or two before the accident, or the “last act” before the accident happens, and is fine for most purposes.
What it isn’t fine for is determining whether an entire class of drivers cause more or less risk as a whole. If a class of drivers routinely drives slower than surrounding traffic, for instance, they will almost never be at fault in an accident. But because they disrupt the general flow of traffic, they will be the root cause of accidents that happen by people pretty reasonably just trying to keep up speed. Those drivers are never at fault, but the roads are a much more dangerous place because of them.
So I look at this data completely differently. There’s almost no way for these results to be true unless they’re driving with a sense of caution that is very disruptive. Like in many things we rely on a certain amount of risk just to keep things going on the roads.
Jim Harrison 01.19.18 at 4:56 pm
What set my priors on this issue was reading Martin Gilbert’s year-by-year three volume history of the Twentieth Century. For each year, Gilbert recorded the number of automobile fatalities. What happens all the time is taken for granted, but Gilbert fought this tendency to ignore the routine. The introduction of autonomous cars will be accompanied by disasters and car companies will lie, cheat, and steel as usual; but it will be surprising if their general use doesn’t save an enormous number of lives.
hix 01.19.18 at 5:11 pm
Definitly good ones too. Also consider the opportunities to reduce congestion and fuel consumption once no humans drive themself anymore.
Doctor Memory 01.19.18 at 5:13 pm
Now that cars don’t need steering wheels or pedals any more
Except that they absolutely still do.
We have, as of right now, a number of interesting demonstrations of “driverless” technology (which often requires a driver to be present), primarily done under controlled conditions in the suburbs of the American west and southwest. (Areas notable for their relative lack of density, generally sunny weather, and wide, well-marked and well-maintained roads.) What these demos all have in common is that you cannot buy them.
GM and Tesla are shipping “driver assist” technologies that are designed to allow hands-off driving on interstate highways, require human intervention to change lanes and do not function on city streets.
The end-game of “full autonomy”, when you actually start drilling for answers, is somewhere between 3 and 10 years off. For those of you not in the tech industry, let me translate that statement: “we have no idea when this will happen.”
If Waymo or GM manage to get one of these gadgets to operate with a decent safety record through a Boston winter, it will be time to start re-considering our licensing programs for human drivers. But I wouldn’t advise holding your breath.
Lord 01.19.18 at 5:31 pm
Some bad drivers learn and become good drivers, but they may not get that opportunity in the future. It may become a leisure activity, bumper cars in controlled environments and video simulations.
Just Another Commenter 01.19.18 at 5:34 pm
Are these autonomous driving systems ready to drive anywhere and at any time that the bad human drivers need them to? If the test cherry-picked routes which the software had mastered, and/or conditions that were optimal for their sensors, then the results might not generalize to driving anywhere at any time.
Moreover, if the routes were designed to carefully avoid entirely categories of situations where the software possibly cannot yet cope (places with irregular signage, possibly hand painted, missing lane markers, ongoing construction, mandatory compliance with directions given by a traffic cop, off-road conditions, unusual crossings, etc.) then this might simply not work for people who have to navigate such zones.
CJColucci 01.19.18 at 5:34 pm
I know anecdotally that the 80-20 rule is true, at least approximately, across a wide variety of areas. Has it been shown to be broadly true in a more rigorous manner, or is it just a very handy rule of thumb?
Cian 01.19.18 at 5:55 pm
John – we have no idea how autonomous cars would perform in real world conditions because they haven’t been tested seriously in them yet. As one research engineer put it – the easy problems that people knew how to solve have now been solved. Now comes the hard stuff.
The hype on this (and AI) is seriously out of proportion to the reality. In addition as far as I can tell the company whose made the most progress on this is Volvo, and they’ve been seriously scaling back their claims.
Mike Furlan 01.19.18 at 6:48 pm
“Frederic Bush 01.19.18 at 3:32 am
I don’t think you can keep iterating the 80/20 rule like that and expect it to hold true.”
You are correct, it is being replaced with the new 99/1 rule.
SamChevre 01.19.18 at 7:08 pm
You can also run the Pareto in the other direction: 4% of the drivers are responsible for 64% of the accidents. Even if the all the rest of drivers are worse than driverless cars, if you could get a significant portion of that 4% into driverless cars it would almost certainly help.
Ben 01.19.18 at 7:54 pm
@Frederic Bush: all segments of the probability density function for the Pareto distribution are self-similar, so to attack Quiggin’s reasoning you’d have to argue why the first 20% of the underlying distribution tracks the Pareto while the following segment of 21%-36% doesn’t
For example: if the relationship of drivers and accidents were that 75% of drivers have one accident, 5% have none, and 20% of drivers have 16 apiece, then the Pareto distribution would model the worst 20% of drivers nicely but would inaccurately assign accidents to the worst 21-36%
Layman 01.19.18 at 8:09 pm
“Quibbling about “who will deliver packages to your door†when we have literally invented a technology that will save more lives per year than a cure for malaria seems incredibly short-sighted.â€
Objection! Assumes facts not in evidence.
Matt 01.19.18 at 8:19 pm
If a class of drivers routinely drives slower than surrounding traffic, for instance, they will almost never be at fault in an accident. But because they disrupt the general flow of traffic, they will be the root cause of accidents that happen by people pretty reasonably just trying to keep up speed.
…
There’s almost no way for these results to be true unless they’re driving with a sense of caution that is very disruptive. Like in many things we rely on a certain amount of risk just to keep things going on the roads.
You don’t need to violate posted speed limits “just to keep things going.” (If you’re referring to people driving under the minimum speed, that’s not following the formal rules either.) The better equilibrium is one where vehicles going over the limit are the outliers. Rigorously speed-limit-respecting robocars should push traffic in general toward that better equilibrium.
(I make no concrete predictions about when robocars will be suited for general use, only about the consequences once they are so suited.)
Omega Centauri 01.20.18 at 3:17 am
You don’t need to violate posted speed limits “just to keep things going.â€
Well in California my speed limit plus 2 or 3, puts me at about one percentile, i.e. 99% of vehicles are passing me. And I have noticed that drivers have an uncanny sense for noticing a car up ahead thats being cautious, then they are willing to take a high level of risk to get ahead of that car as quickly as possible. That could be how these autonomous cars are getting into accidents, other drivers are so desperate to get past them that they throw caution to the wind.
c_haesemeyer 01.20.18 at 4:32 am
Where those “autonomous†GM cars actually autonomous? Or did they have a professional on board intervening, and if so, how often did they have to intervene. Is there any reason to believe results obtained under carefully controlled test conditions would scale? How much money will such an “autonomous†car cost once it actually works in anything else but carefully controlled test conditions, counting both the cost of the vehicle and computing and sensing equipment and the public investment needed to make these things work better, and would this money be better spend some other way improving road safety, ie on public transport? (In other words what is the opportunity cost of the approach you suggest.) How would your suggestion be different from just revoking the human bad driver’s license and making them take the bus and/or taxis, except in the way it gives control over the human’s movements to a major corporation?
c_haesemeyer 01.20.18 at 4:40 am
The idea that autonomous cars, even if they ever work, will reduce the total number of cars significantly, seems to me to be based on a failure to acknowledge the reason why there’s so many cars: because of their twice daily use to transport one individual each. This won’t change if the car in question drives itself (which it will never be able to do, but for the sake of argument …). In fact the hype over autonomous cars will make policies that actually address this issue, ie, public transport, less likely because localities will invest absurd amounts into infrastructure designed to help tiny numbers of sort of autonomous cars function sort of well sort of much of the time.
dax 01.20.18 at 6:09 am
Are bad drivers constantly bad drivers? I was under the impression that the system in the U.S. was to give teenagers licenses prematurely and then let them learn how to drive on the road. So bad drivers will be found disproportionally in the young, but only because they are learning by doing. If you make the young bad drivers self-drive, they will never learn how to drive. Maybe that’s a feature and not a bug, but you could also simply delay the teenagers their licenses.
John Quiggin 01.20.18 at 6:22 am
@36 Responding, I realise I omitted the link to the Bloomberg article. Apologies for that – perhaps I need a robot co-blogger.
To answer the question, the GM cars have so far had a backup driver on board, but those to be used next year won’t have a driver or steering wheel. The test environment is ordinary streets in San Francisco, which is about as challenging as any place I’ve driven in. So, AFAICT, no special investment is required here. The autonomous cars drive on ordinary roads, in with current traffic.
On your final point, society has so far been unwilling to force bad drivers out of cars and into taxis or buses, except after a long string of offences. The robot alternative gives them essentially the same mobility as if they were driving themselves.
I agree that what I’m suggesting won’t involve any fundamental change, in the short run. All it will do is save a lot of lives.
bad Jim 01.20.18 at 7:36 am
I am old, and I have a new car, which makes me especially cautious, and attentive to every transgression. Whenever someone makes a left turn in front of me without signalling I regret not having purchased the artillery accessory. Letting such drivers survive will be detrimental to human progress.
The larger problem, of course, is the pollution these lumbering or extravagant vehicles produce. Cheap oil is good only to the extent that it discourages the most destructive sorts of hydrocarbon extraction, like shale rock, or the XL pipeline; otherwise it encourages or permits consumption which will doom us all.
Okay, it’s good for balance of payments; less money going to the theocratic middle east. Otherwise not wonderful.
Layman 01.20.18 at 10:41 am
Again, there is basically no timetable for actual self-driving cars for consumers. The problem isn’t solved, and no one can say when it will be.
https://finance.yahoo.com/news/pump-breaks-buying-self-driving-car-time-soon-205546606.html
JanieM 01.20.18 at 1:45 pm
Whenever someone makes a left turn in front of me without signalling I regret not having purchased the artillery accessory.
I considered the rear end version, for the tailgaters. Couldn’t afford it.
steven t johnson 01.20.18 at 2:33 pm
It seems to me I read quite a few newspaper notices about arrests for driving without a license. I don’t think US “society” has any problem at all with forbidding bad drivers into taxis and buses, even where there are no buses and most people can’t afford cabs, then arresting them for driving anyhow. “Society” just prefers raising insurance rates on bad drivers in addition to fines for individual infractions.
harry b 01.20.18 at 2:54 pm
This is all nonsense. Everyone knows that 90% of drivers are better than the median driver. (Or is it that 90% of drivers know they’re better than the median driver).
More seriously. I’m a goodish driver — don’t drink and drive, don’t speed, only one accident and that was because a teenager decided to drive his car into me from behind a blind corner at 45 mph, and even then, because I was well below the 25mph speed limit I was able to stop dead before he hit, massively reducing the speed of impact. BUT I want autonomous cars for everyone! Well, especially for me.
c_haesemeyer seems right to me. Is there some response? Though, as a traveller, I would care a lot less about travel times in a car if I were able to use that time to read or work or sleep.
Omega Centauri 01.20.18 at 4:55 pm
I think people are making the assumption that bad drivers are bad because of a lack of skill. More important IMHO is bad attitude, ignoring safety rules, and taking lots of small risks -like passing on a bling curve in light traffic. These drivers are perfectly capable of passing a skills based drivers test, in fact part of the attitude problem is overconfidence in their skills. Also and this is likely due to skill deterioration, the most dangerous group of drivers are the very old.
proportionwheel 01.20.18 at 5:39 pm
I’ll believe that driverless technology has truly arrived when an autonomous car can make it to and from my house, at 1200 ft elevation in Vermont, up several miles of steep and curvy unpaved roads with, obviously, no lane markings or curbs, in the middle of winter. Leaving aside braking and traction issues, which humans learn to anticipate and account for, I have trouble imagining how artificial intelligence even navigates* when the markers are nothing but constantly varying snowbanks, and the width of the usable road varies with the weather. Of course my situation is not typical, but it’s not uncommon across the northern tier of states. Hence the popularity of Subarus, for instance.
*and don’t say it’s all to be done with GPS. Google maps shows one non-existent road within a mile of my place, and fails to show one nearby that’s real and has been there for decades, as well.
Good news bob 01.20.18 at 7:50 pm
Public transport was never a thing…
Omega Centauri 01.21.18 at 12:23 am
proportionwhell @46.
I seem to recall the real issue in those conditions is knowing where the edge of the road actually is. Since often its a deep ditch, and if the snow covers it to the same level as it covers the road you can drive right into it and won’t know its there until your wheels drop in. Then its too late!
Kallan Greybe 01.21.18 at 1:17 am
I just remain shocked that even at a left-wing bastion like Crooked Timber the assumption is still that, for the most part, people will own cars. There is no problem that an autonomous car can solve that can’t be solved at least as well, more cheaply and using current technology, just by investing in public transport and ending the implicit subsidies that we still pay out to people driving cars by not taxing them enough.
Frankly it seems to me there are two things to blame there: a deeply ingrained car culture and the endless hyperbole of a tech-sector which still hasn’t got a clue how to justify the ridiculous amount of resources we give it.
bruce wilder 01.21.18 at 4:07 am
Kallan Greybe @ 40
Thank you.
Personal automobiles entail such a ridiculous profligacy! Not just energy use — though that’s pretty ridiculous — but the use of all kinds of materials, plus the land use implications of all those highways and parking lots! And, urban sprawl!
Doctor Memory 01.21.18 at 5:09 am
Layman: I feel like we’re shouting into the void here. :(
bruce wilder 01.21.18 at 5:24 am
I was curious about the claim that G.M.’s autonomous vehicles were innocent bystanders in all 22 reported accidents — it seems highly improbable that a new technology will produce no consequential mistakes. When has that ever happened?
The most serious injury was to a motorcyclist. The autonomous vehicle was in the middle lane of three lanes on a one-way street in heavy traffic and had started to move into the left lane to take advantage of a gap opening up in the lane to its left, but then changed its little autonomous mind in response to changing traffic conditions and moved suddenly back into the center of the center lane, and smacked a motorcyclist who had been splitting lanes between the center and right lane and had entered the center lane as the autonomous vehicle was pulling out of it.
The claim is made that the police assigned blame to the motorcyclist on the theory that the motorcyclist was attempting to pass on the right when it was not safe to do so. That seems more like an application of a theory of traffic law than a definitive analysis of whether the autonomous vehicle’s sensory boundary, algorithm and signaling behavior was adequate to the situation. Arguably, the autonomous vehicle should not have surprised the motorcyclist. You can say the motorcyclist was taking unnecessary chances, but he hadn’t been smacked by the human drivers immediately to the rear of the Cruise vehicle. It may well be the Cruise moved too fast in following the algorithm that dictates taking an exact center position in the center lane, without signalling its changing “intentions” with the appropriate visible hesitations and pacing of movements that human drivers normally produce and for which other drivers look out. And, if the Cruise can “see” a space opening in the lane to its left and act on that opportunity with appropriate pacing, why couldn’t it also “see” the motorcyclist appear within the center lane to its right? The Cruise was moving at 12 mph and the motorcyclist at a reported 17 mph, which means the motorcyclist was moving relative to the Cruise at the pace of a fast walk.
It would not surprise me in our neoliberal dystopia, if the law was shaped to always blame the human and never the business corporation or its robots. It is pretty routine in many localities to blame pedestrians or bicyclists, even for their own deaths or maimings, and excuse vehicle drivers, provided the driver does not flee the scene and is not visibly intoxicated.
John Quiggin 01.21.18 at 5:39 am
As harry suggests, the Dunning-Kruger effect is strong among drivers. That is, the worst 30 per cent or so of drivers are those most confident in their own skills and (I suspect) most committed to car culture. If they were taken from behind the wheel, that might catalyse the kind of broader change needed to move away from personal car ownership more generally.
But, as I said, previously, I’d be happy with an incremental measure that stopped them from killing other people, and themselves.
Gareth Wilson 01.21.18 at 5:41 am
If it helps, one prediction is that we won’t own driverless cars at all. You’ll just call one to take you to work, then another one to take you home. Solves the parking problem too.
John Quiggin 01.21.18 at 5:51 am
@46 Do you think a driver selected at random from the US population could handle this task?
If not, shouldn’t you level the playing field by allowing the autonomous car to do some preparation: for example, being optimized for snow/ice conditions and building its own map of locations where Google maps are unreliable and
As regards traction and braking issues, the car is closer to the necessary information than a driver getting their information from the steering wheel, so it’s far from obvious that it would do a worse job. After all, ABS has become standard precisely because an automated system outperforms most drivers in handling skids.
bad Jim 01.21.18 at 6:40 am
Even autonomous cars are not a solution to rush hour. Atrios also points out that, unlike cab drivers, they can’t put your luggage into the trunk and take it out again. To a considerable extent we’re quibbling over transportation when the problem is land use patterns.
Perhaps the best thing about my EV is not having to use friction to slow my daily thousand-foot descent, recharging the battery instead. The car has a “hilltop reserve” setting, which keeps it from charging completely at home just to allow such braking.
Ben 01.21.18 at 6:50 am
Self-driving proponents often bring up the metric of saving lives. They shouldn’t. Take into account the *massive* subsidies for self-driving tech and it’s a losing proposition.
Tax breaks, infrastructure, bureaucratic manpower, etc. For *both* development *and* implementation.
Compare the lives saved by investing in mass transit now, through mature processes that would start having an effect in the short term, vs lives that might be saved through unpredictable processes whose implementation time we can’t calculate.
There is no comparison.
proportionwheel 01.21.18 at 6:58 am
@55
proportionwheel 01.21.18 at 7:01 am
@56
So, I apparently don’t know how the tags work. Sorry. Can’t correct, no edit function.
christian h. 01.21.18 at 8:29 am
Gareth @54, where are all those autonomous cars going to be stored given them being autonomous won’t significantly reduce the total number of cars? And how is the transition going to work?
christian h. 01.21.18 at 8:33 am
San Francisco is not a difficult place to drive. And it’s a very restricted location. Can that GM car take you to Santa Cruz if you ask it? It can’t. Because it is not in fact equivalent to a car driven by a human and won’t be for an indefinite time, as layman and Doctor Memory have pointed out.
Also, again, opportunity cost. I seem to vaguely recall it’s supposed to be important.
Layman 01.21.18 at 8:35 am
Gareth Wilson: “If it helps, one prediction is that we won’t own driverless cars at all. You’ll just call one to take you to work, then another one to take you home. Solves the parking problem too.â€
And when everyone in the entire city calls one to take them to work? And all those cars go where, exactly, once everyone gets to work, while they wait for everyone in the city call them to go home? It kinda sounds like the exact same number of cars we have now, with more or less the same parking problem.
Dr. Memory: “I feel like we’re shouting into the void here. :(â€
Indeed. The hype is strong with these ones.
Kallan Greybe 01.21.18 at 9:07 am
Bruce Wilder @ 50
The worst part of the profligacy of car ownership is that the costs aren’t even born by the person who owns the car, to say nothing of the more than 2mn deaths a year that are caused just by people owning and using cars. How much do you want to bet that those costs in turn are disproportionately paid by people on low incomes?
There are plenty of kinds of inequality, but I can’t help but think this is a fairly important one and the debates around autonomous cars still feel like a sideshow.
Ben @ 57
Absolutely right. This is precisely what I find so frustrating about so many political debates, the fact that in most cases we already know how to solve the problems. Take the UK’s housing crisis. During the post-war period there was no housing crisis because the people best placed to diagnose it, local councils, had the power to do something to fix it, build social housing.
Jeremy Corbyn got slammed for “taking politics back to the 70s”, but the reply there is the same reply when it comes to autonomous cars. Forget the self-serving rhetoric of the tech sector. Humanity hasn’t changed significantly in over 100,000 years therefore there is nearly no reason to think that old solutions aren’t perfectly capable of solving today’s problems especially when the problem is the fact that we continue to privilege and subsidise the single most expensive form of mass transit, one with a global death toll in the millions due to accidents and pollution, with absolutely no challenge or debate.
Layman 01.21.18 at 9:57 am
You can find dozens of links to a story on / around 7 November, all of which offer the headline announcements that 1) Waymo has removed drivers from its driverless cars, and 2) Waymo has launched a driverless car service in Phoenix.
Read the actual stories and you will learn that, in fact, the cars still have a Waymo employee in them; that, in fact, they have not launched a car service, but say they will “soonâ€; and that the test they’re running involves a geo-fenced area in the suburb of Chandler, not the city of Phoenix.
Omega Centauri 01.21.18 at 7:26 pm
Layman @62. Assuming the ride service modality prevails, the total number of cars needed should go down. The reason, is that -to the extent that demand can be staggered, the autonomously driven car can on average service more than a single rider. This also would create an economy of scale, which may well be necessary, as the collection of computers and sensors needed by an autonomous vehicle are not cheap. Obviously more cultural tweeks other than just offering rides will be needed. A program to stagger work hours, which to some extent is already being done to spread out demand for freeway throughput, would be a necessary part of this cultural change.
Then, if all cars were autonomous, and in electronic communication with all nearby vehicles, we wouldn’t need traffic lights, or stop signs, and cars could safely travel with very little separation between them. The throughput of a fixed road system would be dramatically improved.
The other development I’ve noticed which is one improvement away from becoming an important factor, is the advancement of electric bicycles. Even a 66 year old like myself could handle a 45mile commute on an electric bike. The remaining obstacle, is the lack of provision of a safe enough route for such bicycles, given the preponderance of cars and trucks.
Gareth Wilson 01.21.18 at 7:58 pm
There’s plenty of traffic between rush hours – every car on the roads then could be replaced by a commuter’s driverless car, freeing up a parking space.
Helen 01.21.18 at 11:26 pm
#asacyclist I fear driverless cars very much, no matter how much logic and science may be thrown at me vis a vis their superiority to human drivers (which of course I also fear.)
It seems to me that driverless cars will constantly be required to solve Trolley Problems which their algorithmic makeup will be inadequate to solve and that I will be the collateral damage.
This may not be a rational fear but it’s going to be a real one for me should these gadgets ever become a real thing on our local roads.
John Quiggin 01.22.18 at 12:01 am
Ben @57 “Tax breaks, infrastructure, bureaucratic manpower, etc. For *both* development *and* implementation.”
Do you have any links/data on this? AFAICT, autonomous cars are supposed to use the existing road network, so I’m unclear what your mean by “infrastructure”. I’m also not aware of any tax breaks, but that doesn’t mean there aren’t any.
Certainly, if autonomous cars require billions in public expenditure, it would be better to put that money into mass transit. The OP is written on the assumption that they simply replace badly driven cars, without any requirement for public expenditure.
Gareth Wilson 01.22.18 at 12:56 am
I’ve driven every day for twenty years, and I’ve never been involved in any kind of Trolley Problem situation.
Omega Centauri 01.22.18 at 1:29 am
Helen@67. Interesting. Now I don’t worry about an autonomous car “solving” a trolley problem, but a minor biking decision today did get me thinking. I came to in suburban intersection alongside a car, we were both going to turn right, and there was a bike lane too. I decided to wait, which I normally wouldn’t -i.e. I would trust the driver to not turn too sharply. But, in this case the car had a driving school sticker on it, and I decided discretion should get the better part of valor. I was guessing autonomous driving SW would not notice such a clue.
Cranky Observer 01.22.18 at 2:13 am
Level 5, and probably Level 4, autonomous vehicles will require networks of sensors in, beside, and above roads communicating to/from vehicles, allowing them to anticipate around corners, acknowledge traffic signals, etc. Similar to ADS-B and TCAS systems in the aviation world except hundreds of times more complex.
One thing that the Big Three and their supply chain have been doing that Tesla, Google, etc have not to any great extent is participating in the multitudinous USDOT, National League of Cities, National Committee on Uniform Traffic Control Devices, ASCE, SAE, etc committees that define codes, standards, and requirements for roadway infrastructure and signalling, with a view to shaping the future requirements for their purposes.
Even the proponents of autonomous vehicles admit that the total infrastructure spending needed in the US to get close to Level 5 would run in the hundreds of billions of US dollars – possibly trillions. As Duncan Black (Atrios) notes this money will just be taken out of the banana stand and spent while other unnecessary programs such as health care and Social Security are chopped due to “not having any money”.
Interesting note: my classmate who attends some of these committee meetings tells me that even relatively poor traffic districts are somewhat willing to spend the money on networked sensor and signaling systems. What they are not willing to do is spend the money for SuperSignals 1.0 in the 2010s only to have the autonomous vehicle manufacturers come back in 2020 with SuperSignals 2.0 which requires a $500 billion “upgrade”. That sort fiscally of responsible thinking doesn’t go over well with Silicon Valley.
christian h. 01.22.18 at 2:25 am
Omega Centauri @65: you are aware that you are suggesting we change the way society works – how work is organised etc – for the sole purpose that we can replace human-driven individual vehicles with not quite as many, but still individual (one point epsilon cars per passenger) vehicles. This just boggles my mind. If reorganising society is that easy, why not do so to reduce the need for and length of trips, increase use of mass transit options etc?
JQ @68: to the extent that ‘autonomous’ vehicles are in fact autonomous this relies crucially on both roadway and roadway marking quality, and on very detailed mapping. Doing the first, and not screwing up the second, would require major infrastructure investment.
John Quiggin 01.22.18 at 6:12 am
Cranky @71 I’m keen to clarify this. My reading suggests
(i) that the sensors required are self-contained in the vehicles, not part of the road network, so the cost would be borne by drivers (or rather, passengers)
https://www.sensorsmag.com/components/three-sensor-types-drive-autonomous-vehicles
(ii) in the medium term, there will be infrastructure savings from more accurate spacing of cars (no tailgating, traffic jams caused by unnecessary sudden braking etc)
http://digitaleditions.sheridan.com/publication/?i=362529&article_id=2649529&view=articleBrowser&ver=html5#{%22issue_id%22:362529,%22view%22:%22articleBrowser%22,%22article_id%22:%222649529%22}
It seems as if you have access to lots of potentially contrary information, so links would be great.
Layman 01.22.18 at 8:43 am
“If reorganising society is that easy, why not do so to reduce the need for and length of trips, increase use of mass transit options etc?â€
Yes, precisely. If we can legislate to 1) end car ownership and 2) spend billions, maybe trillions on mass transit, are not-yet-existent self-driving cars really the way to go? Why not create mass transit systems using, well, mass transit?
I mean, what problem are we trying to solve here?
Joseph Brenner 01.22.18 at 1:48 pm
Layman @74:
“I mean, what problem are we trying to solve here?”
We’re trying to create a drop-in technical fix for the suburban dream.
And people willing to spend 2-3 hours a day commuting will be
willing to increase that to 4-6.
The porta-potty attachments for those high-tech driverless cars
will be interesting. I predict a resurgence in interest in RVs.
But robot cars will be more energy-efficient. Somehow.
(Improving safety over human drivers is setting a pretty low bar
because they suck so badly. One would think that “none of the above”
is an option worth evaluating.)
Trader Joe 01.22.18 at 4:23 pm
I share most of the skeptical views on how quick and how efficient these autonomous vehicles will be. I think there are some limited circumstances where they will make sense but would also expect adoption to be rather mixed. Most polls suggest a great deal of skepticism (as is often the case with something technical that impacts a fairly common behavior)….so it could well be a generation or more before there is enough critical mass of autonomy for the herd immunity of the tech to really have much benefit.
I’d also note with respect to “sharing” that there are a lot of ways to slice that pie. In cities there could be fleets of taxi like vehicles as some note. In suburbs however many families could simply have one car shared between two people – i.e. one spouse could ride into work and send the car back to the other for them to go to work, take kids to school….people would micro coordinate these things among themselves it wouldn’t necessarily require any societal change and even in my simple example could cut vehicle count by 50%.
mpowell 01.22.18 at 6:06 pm
Kallan @ 63: Feel free to run for office on a platform of moving people from driving cars to taking more public transit. I just don’t understand how one gets to the point of believing their preferred policy is both people-oriented and in direct contrast to the very explicit and implicit desires of the vast majority of people.
Yes, you can describe driverless cars as a technical fix for the suburban dream. But this is how people want to live! That should be nearly the entire goal of technology – to make it easier for people to live the way they want to live.
Layman 01.22.18 at 9:57 pm
Trader Joe: “i.e. one spouse could ride into work and send the car back to the other for them to go to work, take kids to school…â€
I wonder, can you trust your school kids alone with a robot car? Or would that be neglect?
That aside, people can carpool now, and most of them could get by with one car by doing that (kids ride the bus to school). People could ride the public bus and have no cars! But, by and large, they don’t. Unless you’re going to legislate away the right for people to own and use cars the way they want to, I don’t see robot cars solving any of those problems.
Layman 01.22.18 at 10:03 pm
mpowell: “That should be nearly the entire goal of technology – to make it easier for people to live the way they want to live.â€
I have to say that this is not at all clear to me. I’m sure that without much effort, you can imagine some ways that some people want to live, where we should not be too quick to develop technology to make it easier for them to do so.
Helen 01.22.18 at 10:29 pm
That’s just looking at things through the USian blinkers, mpowell, which unfortunately is all too common in these fora. My Australian kids are very meh about getting their drivers licences and my neice and nephew have both had stints living in Germany and they were very happy getting about on public transport – because PT worked there!
Noone is talking about banning motorised transport altogether, panic merchants. With vastly improved PT and bike infrastructure you will not be forced to use it but you will enjoy less congestion thanks to others using it.
Kiwanda 01.23.18 at 2:23 am
This guy has been a robocar (as he calls them) enthusiast for quite a long time, and has a lot of general arguments on the “pro” side. His claims of the major savings (in the US):
– 33,000 lives and a million injuries (NIH). Mostly young people, for whom car accidents are the leading cause of death among major categories. Over a million lives/year around the world.
– 230 billion dollars of accident cost (NTSB). About 2-3% of GDP.
– 50 billion hours (or 1 trillion dollars) of people’s time. Around 8% of GDP.
– 50 billion gallons of imported gasoline, replaced with the equivalent of 10 billion “gallons” of domestic-source power plant fuel. Thus eliminating about 12-15% of the USA’s CO2 emissions and nastier pollution.
– A serious reduction in the urban land devoted to the ~600 million parking spaces, estimated to be up to 10% of urban land in many cities.
The arguments re energy savings come from: taxis that can be electric and short range; reduction in congestion; less parking and looking for it; lighter vehicles (because less concern about collisions). The latter requires that the great majority of car be self-driving, of course.
lemmy caution 01.23.18 at 3:12 am
True Driver-less cars would be great but they are nowhere near being available.
People are talking about infrastructure improvements as an alternative or supplement to car sensors :
https://www.freep.com/story/money/cars/2017/06/25/ready-not-self-driving-cars-coming/380845001/
All three of the sensor systems mentioned in that Sensor mag article are currently used in some autonomous cars. There already are LIDAR sensors but they are super expensive and look dorky- it is a roof mounted camera (not in teslas or anything). The sensor mag people are just saying that future commercially available cars will have cheaper LIDAR sensors that will have to be developed.
A big problem with driverless cars is they can’t interpret what people are going to do
https://spectrum.ieee.org/transportation/self-driving/the-big-problem-with-selfdriving-cars-is-people
people interpret the intentions of pedestrians, traffic cops and other drivers all the time. autonomous cars can’t do this. Which is another reason some are pushing for the infrastructure improvements in the hope this can be bypassed.
John Quiggin 01.23.18 at 7:14 am
“people interpret the intentions of pedestrians, traffic cops and other drivers all the time. ”
Referring back to the OP, a lot of drivers do this very badly, because of incompetence or deliberate disregard for others (for example, drivers who deliberately drive through pedestrian crossings, terrorise cyclists etc). On the evidence, these drivers do much worse than autonomous cars.
John Quiggin 01.23.18 at 7:18 am
Following up on various comments, the infrastructure requirement I’ve seen most commonly cited for autonomous cars is that of keeping lane markings in good repair. That’s not free, but I doubt that you could finance much in the way of alternative transport investment by skimping on road paint.
P.M.Lawrence 01.23.18 at 9:38 am
First they came for the bad drivers, but I wasn’t a bad driver so I did not speak out …
christian h. 01.23.18 at 9:57 am
We are talking here not just about repainting lane markings once a year. We are talking about keeping all the detail that goes into the maps these not very autonomous cars must use because they cannot actually drive autonomously unchanged. And then when it snows and covers up those features, oops, all those shiny toys don’t work. And that’s ignoring the fact that it will take fun loving people five minutes to figure out some minor piece of vandalism that’ll make the autonomous cars bricks, have them stop in the road because they “see†an obstacle that isn’t there, go in circles … and then the company owning the cars (or their software) goes out of business or decides to move into terraforming instead.
The things do not currently work outside strictly controlled and delimited environments. Nobody knows if and when they will. The list of practical issues in their widespread use is virtually endless. They do not solve any problem we don’t know how to solve better some other way.
Trader Joe 01.23.18 at 1:14 pm
@78 Layman
“I wonder, can you trust your school kids alone with a robot car? Or would that be neglect?”
I don’t know, that’s probably a matter of personal choice – you trust your kids on a bus, why not a car with locked doors?. Its 100% certain there are more instances of bus drivers who showed up drunk, on drugs or were sexually or physically abusive than there will ever be cars that do the same.
“That aside, people can carpool now, and most of them could get by with one car by doing that (kids ride the bus to school). People could ride the public bus and have no cars!”
In big cities that may be true, in most mid size cities most likely you’ll need to drive several miles to get to a bus/train stop to then take public means. People of course can still do this if they want. I’m not sure I understand the opposition to people utilizing an expensive asset more efficiently….you want two cars and drive them yourself – go ahead. If you want one car (probably electric) autonomously driven why is that a problem? I’m not saying this is tomorrow (or even next 10 yrs) – but if you don’t think its happening you’re missing a mega-trend right before your eyes.
Cian 01.23.18 at 1:56 pm
Following up on various comments, the infrastructure requirement I’ve seen most commonly cited for autonomous cars is that of keeping lane markings in good repair. That’s not free, but I doubt that you could finance much in the way of alternative transport investment by skimping on road paint.
Err no, the problem is the crumbling road surface (not to mention numerous potholes). Though I suppose the ghost lane markings could be a problem also.
Then there are the inconsistent ways that different cities choose to do road markings. The US is allergic to standards – which is a problem for human beings, but toxic for computers.
Lemmy caution 01.23.18 at 3:19 pm
Don’t follow lane markers into a wall like Wiley coyote is something humans handle better than Teslas https://www.google.com/amp/bgr.com/2017/03/02/tesla-crash-video-texas/amp/ giving government the duty to always maintain lane markers (
Even during construction) is a non trivial task
Lemmy caution 01.23.18 at 3:26 pm
I am not arguing safety. When they work they will be safe enough. Even Tesla’s are safe for what they do. They just don’t work At level 5 and they won’t work at level 5 in 3 years or whatever.
Lemmy caution 01.23.18 at 3:39 pm
No autonomous car can take instructions from a traffic cop. That is something bad drivers can do just fine. Bad drivers also generally follow eye contact rules for interacting with pedestrians (they are just dicks about it). I mean eventually they will get there but not anytime soon
Layman 01.23.18 at 3:55 pm
Trader Joe: “I don’t know, that’s probably a matter of personal choice – you trust your kids on a bus, why not a car with locked doors?â€
Currently, if you leave your kids in a locked car that isn’t even moving, you might be going to jail. I’m not talking about a personal trust decision, I’m asking what the law will make of this, e.g. may parents put infants / toddlers / minor children in robot cars alone and send them somewhere, or is this parental negligence?
“In big cities that may be true, in most mid size cities most likely you’ll need to drive several miles to get to a bus/train stop to then take public means.â€
Sure, but that’s a problem of insufficient infrastructure. It could be solved with a bigger fleet of buses covering a greater footprint with more frequency. It’s hard to understand why building up infrastructure for a public robot car service is preferable to building up infrastructure for a better bus service. So far, the answer seems to be ‘personal choice’, but to make robot cars actually work, we’ll have to constrain personal choice anyway.
Layman 01.23.18 at 4:23 pm
This is a great example of the genre:
https://techcrunch.com/2018/01/23/uber-ceo-hopes-to-have-self-driving-cars-in-service-in-18-months/?ncid=mobilenavtrend
Uber’s CEO acknowledges that near-term driverless cars will be nearly useless, guessing that they will work perhaps 5% of the time in one easy test city in 18 months, but taking it on faith that something will change that in 5 years time. This gets translated into the headline that Uber will have autonomous cars operating a ride hailing service in 18 months.
Omega Centauri 01.23.18 at 5:26 pm
Sure people guess the intentions of pedestrians and drivers. Most do it with maybe 90% accuracy. But, many drivers assume that the next most likely action will not happen, so if the human does that, they are not prepared.
Chris Stephens 01.23.18 at 6:17 pm
One of the big benefits of autonomous cars, if enough people use them, is reduced need for parking spaces. The average car sits in a parking space something like 97% of the time. If a car can drop me off, then go pick up someone else, etc. this can dramatically reduce the need for the number of cars and the amount of parking overall.
Cian 01.23.18 at 10:04 pm
One of the big benefits of autonomous cars, if enough people use them, is reduced need for parking spaces. The average car sits in a parking space something like 97% of the time. If a car can drop me off, then go pick up someone else, etc. this can dramatically reduce the need for the number of cars and the amount of parking overall.
Well not really, everyone wants to use cars at the same time. Incidentally when you see somebody in the press suggesting this idea, it’s a pretty strong tell that they’re completely clueless.
Cian 01.23.18 at 10:22 pm
The timelines being proposed for autonomous cars are crazy. For 3 years to be feasible there’d have to be a prototype today that works. Not works in highly controlled circumstances. Because it’s going to take you at least 3 years to take that prototype, work out how to mass produce it, source parts and work out how to make the thing economically and design a car around it. Level-3 is not at that point yet. So we’re more than 3 years away from level-3 being on the roads. Given the status of the prototypes (and the ways in which dates keep being rolled back) I would guess we’re further away than that.
One reason to be skeptical about Google’s approach is that it relies upon mapping _everything_ to a very high level of detail (way beyond google maps), with a far lower tolerance for error than any existing google technology. For their cars to work the maps have to be constantly updated (something they haven’t done in Google Maps). The mapping requirements are technology, time and expertise intensive. It’s hard to see how this would scale and remain cost effective.
Now it’s possible they’ll find ways to automate this away, but it’s unlikely anytime soon (which is another way of saying it may never be solved). In a sense Google have simplied one hard problem (self-driving cars) by replacing it with another (accurate automated mapping).
John Quiggin 01.24.18 at 1:58 am
This thread takes me back to the days when people earnestly defended command-line interfaces against unstable, crash-prone Macs. Or, going back even further, the great
8-bit vs 16-bit debate
https://books.google.com.au/books?id=WYnHD9WSWdAC&pg=PA105&lpg=PA105&dq=8-bit+enough&source=bl&ots=QlVqkAAz4q&sig=KLTOOgGyAhd40wFoly0tXDfRe5U&hl=en&sa=X&ved=0ahUKEwjD-obkv-_YAhVCspQKHZE1B_sQ6AEIgQEwEA#v=onepage&q=8-bit%20enough&f=false
christian h. 01.24.18 at 2:16 am
Only they aren’t Macs. A better analogy is the always ten years away fusion reactor that will solve our energy problems once and for all, or clean coal technology that will very soon arrive I am told.
JanieM 01.24.18 at 3:10 am
JQ @ 98 Or maybe it will be more like these. ;-)
The whole autonomous car topic reminds me of nothing so much as Jared Diamond’s assertion that “invention is the mother of necessity.”
As for Macs, I bought one about three days after they came on the market. The motherboard was flawed and Apple replaced it about a week after I bought the machine. I never had another bit of trouble with it after that in eleven years of use (with a couple of memory and hard drive space upgrades along the way. Were they crash-prone? Who knew!
John Quiggin 01.24.18 at 5:45 am
@100 I also bought my first Mac very soon after they came out, but I had daily encounters with the bomb icon
https://en.wikipedia.org/wiki/Bomb_(icon)
and less frequent, but scarier run-ins with the Sad Mac.
https://en.wikipedia.org/wiki/Macintosh_startup#Sad_Mac
They didn’t change the fact that the Mac was the way of the future. Indeed, within a year of the first workable Windows system coming out, it was just about impossible to find anyone (except Unix fans) defending command line interfaces.
Layman 01.24.18 at 11:26 am
Actually, to me, the self-driving car hype reminds me of nothing so much as the Segway launch, with its attendant claim that Segways would revolutionize transportation.
Trader Joe 01.24.18 at 1:30 pm
@92 Layman
“Currently, if you leave your kids in a locked car that isn’t even moving, you might be going to jail. I’m not talking about a personal trust decision, I’m asking what the law will make of this, e.g. may parents put infants / toddlers / minor children in robot cars alone and send them somewhere, or is this parental negligence?”
I misunderstood your initial point ….I wasn’t assuming sending kids alone. I was assuming a parent or responsible person would “take them to school” much as many parents do every day by driving themselves or car pool (pre-schoolers go to day-care in much the same way – no busses). A agree, I doubt that any law would find it ok to send minors unaccompanied, certainly not in the first 50 years of a new technology, maybe never.
My point remains though that most families would much more easily get by on one car if the autonomous vehichle could be sent empty between the members to be shared. Its certainly possible to car share now, but most do it only from necessity as the logistics aren’t usually worth the relatively low cost of a second car….where the cost of the second car is higher, its more likely people share.
Lemmy caution 01.24.18 at 1:44 pm
Level 5 self driving cars will be great. We just are not 18 months or 3 years or whatever away from them. If they work, people will use them. Christian is right about how some technologies always seem like a breakthrough is coming real soon that never comes. Maybe level 5 cars are one of those or maybe we get them in 10 years
Katsue 01.24.18 at 3:25 pm
How would there be reduced need for parking spaces? The whole point of autonomous cars is to sell cars to people like myself who don’t want to learn to drive. If they actually work, then there will be even more cars on the road.
Layman 01.24.18 at 5:32 pm
This seems on topic.
https://www.wired.com/story/tesla-autopilot-crash-dui/
Al 01.25.18 at 1:55 am
Following the logic of that classic dsquared post on Milton Friedman: no, let’s not suppose that. Let’s suppose instead that each of the self-driving cars in these crashes is operating with the same software (and thus not like tossing a coin randomly). Then the probabilities of all of these crashes is unlikely to be independent – if there are particular failure routes that humans exhibit, they’ll line up whenever the variables match those human failure routes. And similarly, if there are particular failure routes that the self-driving software exhibits, then they’ll line up whenever the variables match those self-driving failure routes. As such, if the GM President can be trusted and all 22 crashes were human-caused, then this is *either* (1) evidence humans are on average more dangerous at driving than self-driving cars, *or* (2) evidence that whatever the failure route of self-driving cars (and there will be one, no system is fool-proof), the self-driving cars just happened not to be subject to it during these test runs.
Now, it’s possible that the companies have gathered so much data now as to make any safety concerns unlikely. However, this is a complex, empirical question. If we’re looking for a fat tail, simple binomial probability reasoning is not at all helpful.
(As it happens, my prior is to say self-driving cars are probably safer than regular ones… but that doesn’t mean we should accept faulty mathematical reasoning to prove it!)
Omega Centauri 01.25.18 at 3:32 am
Katsue: The great white hope with respect to parking, is predicated on a change of the ownership model. The assumption is that fewer cars overall would be in circulation, and that a frequent trip mode would be the car picks up its passengers and drops them off, then goes on pick up the next passenger, so the vehicle carries several riders between parking. Whether that happens remains to be seen….
No, people don’t all drive at the same time. Where I live in Bay area the morning commute is spread between 5am to almost 9am, and few commutes take four hours. Some leave early, and some leave later. The spread is largely caused by people trying to avoid the worst congestion.
So while there are busy periods, and not so busy periods, these busy periods are long enough that an automatic taxi could complete more than one trip per busy period. Now maybe for small towns, where traffic jams are rare, the commute spike is much sharper, but much of the population is now in the megacities.
TM 01.25.18 at 8:39 am
108: It seems that this effect could happen *if* authorities managed to limit the total number of cars on the road. Then algorithms could optimize the efficiency of the network by spreading out demand for driving services over the available capacity. But without a mechanism limiting that capacity, I would expect that the number of cars would be no less than it is now.
John Quiggin 01.25.18 at 10:46 am
AL @ 107 Nice point, but I’m not sure how far it gets you. It’s certainly consistent with the idea that the failure route for autonomous cars exists, and that it will be encountered with some positive probability p, while the failure route for humans will be encountered with probability q (in both cases, these probabilities are for environments similar to SF). But however you formulate the model, after observing 22 instances of human failure and zero of autonomous, you ought to have pretty high confidence that p
Katsue 01.25.18 at 1:15 pm
@Omega Centauri
General Motors and Volvo are not researching driverless cars in the hope that, after doing so, there will be lower demand for their products.
Cian 01.25.18 at 5:03 pm
Omega Centurai: It’s not the morning commute period that you worry about, but when people are traveling. Most places have a fairly narrow peak of maybe an hour when most people are traveling. And any ‘hire car’ company would need to have sufficient capacity to serve that period of time. That extra capacity would be sitting in a lot the rest of the day.
Trains, metros and buses have been dealing with this problem for over a hundred years, but obviously nobody is interested in learning from such antiquated technologies.
Al 01.25.18 at 5:13 pm
@110
Seems to be a formatting error at the end there, I assume you were saying “high confidence that p is less than q”. But if we’re dealing with a fat tail here, this could be just like someone claiming in 2005 “subprime mortgages aren’t a risky bet”. So it’s only high confidence *under the assumption of a particular normally distributed model*.
Cian 01.25.18 at 5:15 pm
John, I don’t think you can safely draw any conclusions on safety from a few carefully controlled trials. Particularly when you’re drawing conclusions about safety for a technology that literally DOES NOT EXIST. Fully autonomous cars don’t exist in prototype form. Given the current state of the technology we’re not close to fully autonomous cars IN PROTOTYPE FORM. In fact, given the current state of autonomous cars we can’t even confidently state if autonomous cars will exist in our lifetimes. AI is filled with technologies that have been perpetually 5 years away since the 70s.
AI researchers are notorious for over promising and under delivering. Currently autonomous cars are following that trajectory.
Fun fact about bayesian/neural networks. Nobody really knows when, or why, they will fail. Which makes their use for safety critical systems interesting. Also makes reasoning about their likely safety affects kind of tricky I would think.
John Quiggin 01.25.18 at 10:12 pm
@113 Sorry for typo. On the fat tails point, are you thinking about a failure route in which large numbers of autonomous cars crash at once? That would seem to be consistent with your rejection of independence. But I’m not seeing how that works. Once you have independence, the Central Limit Theorem gives you normality in large numbers of trials.
To various commenters, I’m aware that hype always outruns reality. Still there’s a huge gap between the industry view here (median projection 2021)
https://venturebeat.com/2017/06/04/self-driving-car-timeline-for-11-top-automakers/
and the comments above.
Priest 01.25.18 at 11:10 pm
I think the gap is partly attributable to some fuzziness in the discussion – the automakers are (largely) talking about Level 3 & 4 “self-driving”, whereas a lot of the skepticism seems directed at how soon (if ever) Level 5 is achieved. For my purposes, at least, “self-driving” = my status in the vehicle is as a passenger with no liability for the performance of the vehicle – no insurance needed, no risk of arrest for DUI, etc. By that criteria, 2021 seems. . . optimistic.
Suzanne 01.26.18 at 12:56 am
21: “Cars kill roughly 1.2 million people per year worldwide. Something like 90% of those are preventable. Quibbling about “who will deliver packages to your door†when we have literally invented a technology that will save more lives per year than a cure for malaria seems incredibly short-sighted.”
Not to mention elderly or handicapped people in suburban areas who are largely trapped in their homes when they can no longer drive, having to rely on not terribly efficient shuttle or bus services, relatives, and friends. Many old people keep driving longer than they should because of the limited options available to those who can’t drive themselves. (Should they have better options? Yes. But they don’t.)
Yes, I would prefer much more and better public transportation and fewer cars on the road. Until that happy day arrives, I hope the industry is right and the naysayers are wrong.
Layman 01.26.18 at 1:22 am
@JQ, that article includes as its most testable prediction that of Tesla and Elon Musk, of having fully driverless cars on the highway by 2017, and a car that would drive itself unaccompanied from NY to LA before the end of 2017. We can agree that was total bullshit, right? Is there any reason to think the others are less so?
Faustusnotes 01.26.18 at 2:15 am
These autonomous cars are going to be disgustingly dirty. With no driver and multiple unconnected users they will be treated like a rubbish dump. Very soon anyone who can afford it will buy and use their own.
Cian 01.26.18 at 2:08 pm
John – the industry view (by which I mean the CEO view) are generally the ones leading the hype. Not sure why you think it would be any different for self-driving cars.
3 years seems ridiculously optimistic given that nobody has trialed a prototype in realistic driving conditions yet (let alone worked how to build these things within a reasonable budget). That number seems driven more by PR needs, than anything particularly technical.
As for the push back that you’re receiving. Well much of it seems to be coming from people like myself who work in the tech industry. If you work in the tech industry you’re used to the following:
+ Worldchanging technologies that turn out to be vaporware (there are a lot of these).
+ Worldchanging technologies that turn out to be pretty crap/underwhelming (lot of these).
+ Worldchanging technologies that eventually leap out many years later (lot of these).
+ Incremental improvements that are fine, but incredibly overhyped (not as common as they should be).
With automated cars you have two tendencies that should make anyone skeptical. First of all it’s AI. Secondly it’s a technology that needs to solve a wicked problem (in the West Churchman sense). Embedding technologies in complex social spaces is hard and the tech industry usually does a crap job of solving it. In fact the tech industry usually isn’t even capable of spotting that a wicked problem exists.
Cian 01.26.18 at 2:11 pm
Also the assumption that self-driving cars will be safer relies upon assumptions rather than evidence. We really don’t have any good evidence with which we can make any predictions about these things. There is an exponential difference between a controlled trial and the real world. They may be safer, but it really isn’t the slam dunk that people seem to be assuming.
c_haesemeyer 01.27.18 at 4:21 am
Measuring safety by who is legally “at fault†is highly questionable. When a Tesla on “autopilot†didn’t see a truck and ran into it, the NTSB concluded that the truck driver was at primary fault for taking the Tesla’s right of way. Driving is a social exercise, and among human drivers it commonly accepted that technical violations of your right of way that force you to slow down or switch lanes occur regularly – when vehicles from smaller roadways cross bigger roadways, or turn into them, on left turns etc. You may think this is bad driving, but if you try to come with a way to reduce fatalities you must take this into account. (In fact under a plan that introduces only a small number of autonomous vehicles into a much larger number of human controlled ones by replacing a subset of the worst drivers only, the number of fatalities could increase even if no autonomous vehicle is ever at fault.)
Omega Centauri 01.28.18 at 1:34 am
Priest @116 DUI? I read about the Tesla DUI on the Bay Bridge. A very drunk Tesla owner tried to use autonomous mode to drive home. The software requires that the driver is still present and attentive, so if it doesn’t detect driver inputs over a predetermined length of time, it goes into red-hands of death mode -turns off autonomy, and stops. The police investigates a stopped Tesla on the bridge with a passed out driver with 2x the legal limit. He was hauled off to jail and his car impounded. So until we have full level-5, you won’t be able to use autonomy to drive you home legally.
Divelly 01.28.18 at 5:10 pm
@15
An interview with a GM engineer on Click & Clack:
C&C :
“Why don’t car clock work?”
GM:
” Not only do they work, they are the most reliable system in every GM product!”
Comments on this entry are closed.